Re: RMSD metadynamics

From: Jérôme Hénin (
Date: Thu Feb 16 2012 - 02:39:04 CST

Hi Simone,

Yet another option: define a single variable that is the difference
between RMSD with respect to one structure and RMSD with respect to
the other one, and get a 1d PMF. But I recommend the 2d option, since
that will reduce resolution problems where very different
conformations give the same value of the collective variable. 1d
methods will work well only if your protein is intrinsically not too
flexible and will move along a well-defined pathway without distorting
too much. That is often hard to know in advance, and not very likely.


On 15 February 2012 23:38, Giacomo Fiorin <> wrote:
> Hi Simone, no, I meant two structures with respect to which the RMSD can be
> computed.  The way I understood your question is that you know already two
> alternative conformations of your protein, "start" and "end", and you want
> to study the transition between the two.
> For this reason I suggested that you define two RMSD variables, with respect
> to the "start" and "end' structures, respectively.  Then you use both these
> variables with metadynamics, getting a 2d PMF.
> If you absolutely want to use just one variable, then you can compute the
> vector difference between the "start" and "end" structures, and use that to
> define the "eigenvector" variable instead of RMSD.  This way you would be
> studying a transition path that is a linear interpolation between "start"
> and "end".  Sometimes this is also referred to as "path collective variable"
> in other metadynamics simulations (in the typical case where the linear
> interpolation is considered).
> What you wrote that at the end the RMSD is e.g. 10, simply means that you
> are pulling the RMSD variable away from zero (or the protein away from the
> "start" structure).  This is essentially the inverse of targeted MD.
> Instead, in a metadynamics simulation the end configuration depends only on
> when you stop the simulation, and shouldn't matter once you have
> reconstructed the PMF.
> If, in addition of wanting to use one variable, you want to finish the
> simulation in the "end" structure, initialize the protein in the "start"
> structure, define one RMSD with respect to the "end" structure, and pull
> this variable towards zero with a moving harmonic restraint.  This is
> essentially, targeted MD.
> On Wed, Feb 15, 2012 at 4:59 PM, simone bardwell <> wrote:
>> Thanks for this Giacomo and for the tips regarding searching the user
>> list.
>> If I could ask you to clarify one further point, when you say 'If you need
>> two structures (start and end), you can either define two RMSDs, or use a
>> combination of the two values', do you mean at the start the RMSD is 0 as it
>> overlays with itself and then at the end the RMSD is, for example, 10? Sorry
>> if I've missed the point here.
>> On Wed, Feb 15, 2012 at 7:48 PM, Giacomo Fiorin <>
>> wrote:
>>> Hello Simone, the RMSD is with respect to one structure.  You provide
>>> this structure via e.g. a PDB file, define the atoms to be fitted onto it,
>>> and the RMSD will be computed at every time step, on the same atoms.  If you
>>> need two structures (start and end), you can either define two RMSDs, or use
>>> a combination of the two values.
>>> Searching the mailing list archives a bit more may have provided this
>>> thread from less than two weeks ago:
>>> where each of the answers can certainly help you.
>>> The message referred to ABF, but the difference with metadynamics is just
>>> using "metadynamics" instead of "abf".  My general advice is, search for
>>> answers and/or example regarding a specific variable, because most of the
>>> times you can use on it any method you want.  So searching for
>>> "metadynamics", "abf" or any other method will only narrow down your
>>> results.
>>> There is an example of input file, right in the relevant section of the
>>> NAMD user's guide:
>>> The example is to apply an harmonic restraint to a couple of distance
>>> variables.
>>> To adapt it to one of the 17 other types of variables, as I said try to
>>> search for that variable type first.  Once you defined your variable,
>>> applying metadynamics or one of the other methods is pretty straightforward.
>>> Bests
>>> Giacomo
>>> On Wed, Feb 15, 2012 at 2:25 PM, simone bardwell <>
>>> wrote:
>>>> Hello
>>>> I have a couple of questions regarding setting up
>>>> metadynamics calculations in NAMD. I have looked though the archive user
>>>> list but have not found a response to my questions.
>>>> I would like to use RMSD as a CV for a metadynamics calculation. I would
>>>> like to be able to specify a start and end structure and then a set of atoms
>>>> to use to fit the structures together and then measure the rmsd of a
>>>> different region. I have read the Manual however am not sure if the region
>>>> being specified is for the fit and will also be the region to measure.
>>>> My second question is whether anyone has any practical examples of RMSD
>>>> metadynamics input files they would be willing to share. I have searched for
>>>> tutorial examples for these types of calculations however with no success.
>>>> To be able to refer to an example would be really helpful.
>>>> Kind regards,
>>>> Simone Bardwell

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:13 CST